Dependence of Working Memory on Coordinated Activity Across Brain Areas

Abstract

Neural signatures of working memory (WM) have been reported in numerous brain areas, suggesting a distributed neural substrate for memory maintenance. In the current manuscript we provide an updated review of the literature focusing on intracranial neurophysiological recordings during WM in primates. Such signatures of WM include changes in firing rate or local oscillatory power within an area, along with measures of coordinated activity between areas based on synchronization between oscillations. In comparing the ability of various neural signatures in any brain area to predict behavioral performance, we observe that synchrony between areas is more frequently and robustly correlated with WM performance than any of the within-area neural signatures. We further review the evidence for alteration of inter-areal synchrony in brain disorders, consistent with an important role for such synchrony during behavior. Additionally, results of causal studies indicate that manipulating synchrony across areas is especially effective at influencing WM task performance. Each of these lines of research supports the critical role of inter-areal synchrony in WM. Finally, we propose a framework for interactions between prefrontal and sensory areas during WM, incorporating a range of experimental findings and offering an explanation for the observed link between intra-areal measures and WM performance.

Keywords: brain disorders; causal manipulation; oscillation; synchrony; working memory.

FIGURE 1

Summary of studies examining synchrony between areas during working memory (WM). Brain schematics of the monkey (left) and human brain (right), and areas recorded (green) in studies reporting measurements of synchrony between areas. Gray arrows indicate areas recorded in the same study, labeled with the frequency band in which WM-induced changes in synchrony between the areas were reported.

FIGURE 2

Synchronized activity between brain areas reflected the content of WM. (A) Phase-phase locking (PPL) between PFC (FEF) and temporal cortex (IT) encoded the identity (top) and location (bottom) of the sample object during a delayed-match-to-sample task [adapted from Rezayat et al. (2021)]. Heatmap shows the difference in PPL between conditions (different object identities or locations) across time and frequency. (B) LFP-LFP coherence between PFC and parietal cortex encoded the identity (top) and location (bottom) of the sample object during the delayed-match-to-sample task [adapted from Salazar et al. (2012)]. Heatmap shows the difference in coherence between conditions (different object identities or locations) across time and frequency.

FIGURE 3

A framework describing prefrontal-visual interactions and the functional role of coordinated oscillations during WM. Description moves counter-clockwise from upper left. WM-dependent spiking activity: spiking activity within PFC (red) represents the content of WM. This activity is sent from PFC to visual areas via direct projections (red projection), recruiting sensory areas. WM-induced distant oscillation: within visual areas (blue), the top-down WM input drives an αβ-frequency oscillation. Oscillation-dependent spiking activity: the combination of this αβ oscillation and a neuron’s sensitivity to sensory input will determine its spike timing relative to the local αβ oscillation. Spikes are sent from visual areas to PFC (blue projection). WM-induced local oscillation: WM activity within PFC also drives an αβ oscillation within PFC, which will be phase-locked with that in visual areas. Oscillation gates input efficacy: the phase of the αβ oscillation within PFC will gate the efficacy of visual input, providing a mechanism to preserve the information contained in spike timing relative to the oscillation. Visual inputs to PFC target visuomotor neurons, and evoked activity in PFC will in turn guide behaviors (for example, eye movements), with the net result that incoming stimuli matching the content of WM are more likely to influence behavior. This model is based on results that have been reported in one or more visual areas including V4, MT, and IT, for spatial or object WM (see text for references).